MotionLint reviews the design and motion of your running web app. A vision LLM catches what your coding agent can't see; a deterministic linter measures what it can't feel. In your terminal, in CI, or inside Claude Code.
Coding agents read JSX and CSS — they're blind to what users see, click, and watch animate. MotionLint captures the running app and hands ranked, cited findings back to the loop.
Both chips animate at their real timings — the difference between the finding and the fix is something you feel, not read.
motionlint review captures multi-viewport screenshots and 50ms frame bursts after every interaction, then returns ranked findings across 12 UX dimensions — each with a suggested fix.
motionlint audit measures every animation in the browser and lints durations, easing, stagger and exits against encoded standards. No API key, no LLM — plus SARIF and exit codes for CI.
motionlint tune opens your real animations in a live tuner — sliders for duration and delay, Emil easing presets, instant replay — and exports the diff for your agent to apply.
A native MCP server for Claude Code, Cursor, or any MCP-aware client — findings land next to the code being written, not in a separate dashboard. Validated at 100% recall, 0% false positives on a 24-fixture stress test.
The linter checks real, machine-checkable constants — every finding cites the exact rule and value it was measured against.
Encoded from Emil Kowalski's animation standards. The same constants power the linter's findings, the tuner's presets, and the suggested fixes — one source of truth.
The audit runs fully offline — no account, no API key. For vision review, bring Anthropic, OpenAI, Google, or a local Ollama model.
MIT licensed · TypeScript · runs on Chrome via CDP